Build & sell

Jev and TypeSafe AI Explained

Understand the launch claims, free output tokens, confidence scores and a practical first business trial.

5min read
7sections

Jev is TypeSafe AI’s model for structured decisions inside software. This guide explains the launch claims, what “free answers” actually means, and how to test whether the model fits a business workflow. Its usefulness depends on your task, your data and the rules you build around its output.

What Jev does

Instead of asking for an essay, you supply context and narrowly defined questions. Jev returns typed results that an application can use: a selected option, a score, or a truth estimate. TypeSafe calls this a System One model. The documentation recommends dividing complicated judgments into focused questions, then combining the answers in code. Several questions can be evaluated independently in one call.

For example, a support workflow could ask which department owns a ticket and whether the customer is reporting an urgent problem. The result can change the ticket’s route without requiring a person to read a generated explanation first. That is a useful place to explore Jev; it is not a promise that the model replaces every capability of a general chat assistant.

Read the model introduction

Who built it

TypeSafe’s founder Diogo Almeida worked on the research behind ChatGPT. The company describes him as a co-inventor of RLHF and InstructGPT. Its team page lists backgrounds including OpenAI, Google Brain and Meta/FAIR. “Research behind ChatGPT” is a more precise description than calling someone a founder of ChatGPT.

Meet the TypeSafe team

What the speed and cost claims cover

The September 15 launch presents very large speed and cost advantages for structured System One workflows. TypeSafe’s homepage reports 193.6 times faster and 444.6 times cheaper in its workflow evaluations. These are company-published comparisons, not a guarantee of those gains on every task or against every AI model.

The evaluation site covers four examples: security incidents, agent trace observability, invoice processing and customer service. Reference answers come from a model consensus rather than independently labeled ground truth. The launch post also explains that its demonstration uses short input that favors the new approach. Treat the headline figures as a reason to investigate, then measure the workload you actually need.

Inspect the workflow evaluations

Read the launch evidence and its qualifications

Free, for your email

Get the full setup steps + the copy-paste setup prompt

The rest of this guide (4 more sections) opens right here. The setup prompt, one paste into Claude that sets it all up, comes by email only. Use your real address.

Unlocks: What is free and what is paid · Confidence is useful but is not an accuracy guarantee · Run a small useful trial · A practical first project

50,000+ follow where these guides come from. No spam, reply “unsubscribe” any time.

Read next